On 5 September 2026 AcuiQ measured its database against eleven stated norms and published the result. The dataset at the foot of this page is that measurement: one row per rule, each carrying the authority the rule derives from, the date a person last read that authority, and the share of the corpus that deviates from it.
The rules were already there. AcuiQ has been dropping animal studies at ingest since August, refusing to link a herb whose Latin name matches two catalogue rows, and reporting points whose safety text forbids the condition the same row is indicated for. What none of those rules said was where their norm came from, or how much of the corpus broke them. They are one register now, and this is its first reading.
A count is not a rate
The audit used to report 41 self-contradictory treatments. That number climbs when the corpus grows and falls when it is pruned, and neither movement means the data improved. It now reports 41 out of 6,156 treatments carrying any warning text, which is 0.67%.
Every rule declares its own eligible set, and both numbers are published. A deviation rate without its denominator is a claim about the literature that nobody can check, and this is a health surface. If AcuiQ will not publish the denominator, it should not publish the rate.
Three kinds of norm
A rule's kind decides what evidence is allowed to move it, and the three kinds do not behave alike.
Internal rules ask whether the database contradicts itself. No outside authority is wanted. A point whose caution names the condition it is listed as treating renders both blocks on the same page, one above the other, and a reader cannot act on it. That row is wrong on its own terms, whatever any standard says. These are the rules AcuiQ is closest to clean on: all five sit under 2.4%, and the worst of them is 147 treatment rows shadowed by a same-code sibling, where one Latin binomial carries two safety texts and the lookup keeps whichever it read last.
Documentary rules ask how the record is written: which code scheme, which binomial, which identifier. The norm comes from a named body and never from a tally, because here the majority of the published corpus can be wrong and is. Most papers using auricular codes number the points anatomically. This catalogue numbers them alphabetically. The prefixes are the same in both, so a code crosses the boundary without its scheme attached, resolves to a real but different point, and nothing anywhere reports an error. Counting how many papers use the anatomical numbering would tell you about publishing habit and nothing about which reading is correct.
Empirical rules ask about the world: does this prescription treat this condition. Here the count is the evidence. Unrelated groups arriving at the same prescription is the strongest thing anyone can offer, and a point list asserted by a committee with no trials behind it is the weaker claim, not the stronger one. AcuiQ takes that position and the register encodes it. What the position requires is that the sources actually be independent, and that is where this corpus needs watching.
The register
| Rule | Kind | The norm | Deviation | Violations / eligible |
|---|---|---|---|---|
warnings/self-contradiction | internal | A treatment does not caution against a condition it is indicated for. | 0.67% | 41 / 6,156 |
warnings/protocol-conflict | internal | A protocol does not prescribe a point whose caution names what the protocol treats. | 0.44% | 24 / 5,462 |
warnings/pregnancy-gestational | internal | A point forbidden in pregnancy is not prescribed for a pregnancy meant to continue. | 0.64% | 35 / 5,462 |
indications/inverted | internal | A contraindication is never filed as a symptom the treatment addresses. | 0.00% | 0 / 39,813 |
treatment/single-safety-profile | internal | One code carries one safety text. | 2.38% | 147 / 6,170 |
points/code-resolves | documentary | Every stored point reference resolves to a catalogue row. | 0.19% | 58 / 30,043 |
points/protocol-has-treatments | documentary | A protocol reaching the matcher prescribes at least one point. | 0.00% | 0 / 5,468 |
herbs/binomial-unambiguous | documentary | A Latin binomial identifies exactly one catalogue row. | 18.77% | 751 / 4,001 |
sources/verifiable-identifier | documentary | Every protocol cites a resolvable identifier, not a citation string. | 23.05% | 1,260 / 5,466 |
sources/human-subjects | documentary | Every protocol comes from human clinical evidence. | 0.13% | 2 / 1,536 |
evidence/independent-corroboration | empirical | A symptom AcuiQ answers rests on more than one independent source. | 67.29% | 3,053 / 4,537 |
What the documentary rules found
Point codes resolve well. Of 30,043 point references stored across the corpus, 58 have no catalogue row behind them, which is 0.19%. Most of the remainder are ashi points, which are tender spots found on the patient and have no code by definition.
The herb catalogue is the opposite. 18.77% of its 4,001 Latin binomials carry more than one row, so the name identifies a species and the database cannot say which row it means. That is not a data error waiting to be pruned. It is the reason AcuiQ links no herb protocols at all: picking a row would publish a guess with a citation attached, and the guess would be about which safety text applies.
23.05% of protocols cite a reference that no public resolver dereferences, mostly Chinese-language journals indexed only in CNKI. 1,260 of 5,466 rows. The underlying work may be sound and the extraction may be faithful. Nobody outside can check either.
The number that matters
67.29% of the conditions AcuiQ answers rest on a single source. 3,053 of 4,537 conditions that carry any protocol at all trace back to one reference. Across the whole corpus there are 1.28 protocol rows for every distinct source.
A source is not always a paper, and the difference matters more than the headline rate does. Of those 3,053 conditions, 2,175 rest on one published study, 683 on one clinical text or case report, and 195 on one website. 192 of them – 673 protocol rows – come from a single page at sacredlotus.com. Two more come from flashcard sites. Those rows are labelled on the page that serves them, because AcuiQ resolves what a reference is before it says the word protocol, and a website is not a trial. They are still there, and they still count toward a number a reader might read as evidence.
The page for such a condition still shows a protocol count, and a reader can be forgiven for reading that count as agreement. It is not. Part of this corpus once stored one paper's single prescription as 252 rows, fanned across ages the paper never reported, and every one of those rows passed every check that looks at rows one at a time. Counting distinct sources is what counting rows cannot do, which is why the rule counts sources.
What a source count cannot see
A distinct reference is a publication, not an independent group. The same authors publishing twice read as two here. A trial reprinted in a second journal reads as two. Vickers and colleagues reviewed 252 controlled trials of acupuncture in 1998 and found that every trial from China, Japan, Hong Kong and Taiwan reported a positive result, which is the shape a corpus takes when its sources share more than a subject. Distinct-reference counting cannot see any of that, so 67.29% is an upper bound on corroboration and never a floor.
Tightening it needs author and affiliation, which AcuiQ already resolves for part of the corpus while checking whether a study was run in people. That is the next entry in the register rather than a footnote on this one, and when it lands the rate will move in one direction.
What happens next
The register grows by entries. Adding a norm now costs a record stating what it rests on and what it is measured over, not a script, and the next ones are already named: whether a paper declares the point-location standard it followed, and whether a registered trial ever published its results. Each reading is published with its date, so the rates can be compared with the previous one rather than read as a verdict.
AcuiQ indexes what other people published and checks none of it clinically. The register does not change that. It says which of AcuiQ's own claims about its data can be checked, by whom, and against what.